auto_annot_Smillie2019_with_Martin2019_Type¶
import besca as bc
import pkg_resources
Specify folders where .h5ad files are found and their names.¶
The datasets that are already annotated and should be used for training. If you only use one dataset please use list of one.
# the path to the datasets
train_dataset_paths = [pkg_resources.resource_filename('besca', 'datasets/data')]
#the names of the h5ad files
train_datasets = ['Martin2019_processed.h5ad']
The dataset of interest that should be annotated.
test_dataset = 'Smillie2019_processed.h5ad'
test_dataset_path = pkg_resources.resource_filename('besca', 'datasets/data')
Give your analysis a name.
analysis_name = 'auto_annot_Smillie2019_with_Martin2019_Type'
Now specify parameters¶
Specify column name of celltype annotation you want to train on.
celltype ='Type'
Choose a method:
- linear: Support Vector Machine with Linear Kernel
- sgd: Support Vector Machine with Linear Kernel using Stochastic Gradient Descent
- rbf: Support Vector Machine with radial basis function kernel. Very time intensive, use only on small datasets.
- logistic_regression: Standard logistic classifier iwth multinomial loss.
- logistic_regression_ovr: Logistic Regression with one versus rest classification.
- logistic_regression_elastic: Logistic Regression with elastic loss, cross validates among multiple l1 ratios.
method = 'logistic_regression'
Specify merge method if using multiple training datasets. Needs to be either scanorama or naive.
merge = 'scanorama'
Decide if you want to use the raw format or highly variable genes. Raw increases computational time and does not necessarily improve predictions.
use_raw = False
You can choose to only consider a subset of genes from a signature set.
genes_to_use = 'all'
Read in all training and the testing set.¶
adata_trains, adata_pred, adata_orig = bc.tl.auto_annot.read_data(train_paths = train_dataset_paths,train_datasets= train_datasets, test_path= test_dataset_path, test_dataset= test_dataset, use_raw = use_raw)
This function merges training datasets, removes unwanted genes, and if scanorama is used corrects for datasets.
adata_train, adata_pred = bc.tl.auto_annot.merge_data(adata_trains, adata_pred, genes_to_use = genes_to_use, merge = merge)
Train the classifier.¶
The returned scaler is fitted on the training dataset (to zero mean and scaled to unit variance).
classifier, scaler = bc.tl.auto_annot.fit(adata_train, method, celltype)
Prediction¶
Use fitted model to predict celltypes in adata_pred. Prediction will be added in a new column called 'auto_annot'. Paths are needed as adata_pred will revert to its original state (all genes, no additional corrections). The threshold should be set to 0 or left out for SVM. For logisitic regression the threshold can be set.
adata_predicted = bc.tl.auto_annot.adata_predict(classifier = classifier, scaler = scaler, adata_pred = adata_pred, adata_orig = adata_orig, threshold = 0.1)
Write out metrics to a report file, create confusion matrices and comparative umap plots
%matplotlib inline
bc.tl.auto_annot.report(adata_predicted, celltype, method, analysis_name, train_datasets, test_dataset, False, merge, use_raw, genes_to_use, clustering = 'leiden')
import scanpy as sc
sc.pl.umap(adata_predicted, color=[celltype, 'auto_annot'])
sc.pl.umap(adata_predicted, color=[celltype, 'auto_annot'], legend_loc='on data', legend_fontsize=6)
adata_train